Estimation of forest above‐ground biomass in Guangxi, China, by integrating forest age and stack learning

Abstract

With the implementation of large‐scale ecological restoration projects, Southwest China has become one of the fastest forest growth areas in the world in terms of vegetation cover and above‐ground biomass (AGB). It is expected to be a potential area for achieving the carbon neutrality target in China. Accurate estimation of forest AGB is becoming an increasingly urgent necessity for carbon neutrality and forest management. However, due to the complex geological background, there is significant uncertainty in estimating forest AGB in the southwestern region, which generally results in underestimating carbon sinks from forest restoration. To address the issue, we propose a method by incorporating forest age information and stack learning technique to estimate forest AGB. Based on remote sensing, forest inventory and in situ forest biomass data, three fundamental methods (Multiple regression, Random forest and Support vector machine) are employed and compared to build the AGB estimation model with vegetation indices, texture feature factors and forest age. Optimal basic models are further enhanced by integration learning to improve the estimation performance and then applied to the study area Guangxi to obtain regional AGB information of different forest types. The results show that: (1) forest age plays a vital role in reducing the uncertainty of AGB estimation. By incorporating forest age information, R2 of AGB estimation is improved by 0.07–0.27 and RMSE is decreased by 16.35%–47.47% for different forest types; (2) with R2 value >0.78, random forest model outperforms support vector machine and multiple linear regression models. Compared with the single optimal model, integration model by stack learning further enhances R2 of estimation by 0.02–0.03 and decreases RMSE by 5.20%–14.89%. (3) The total forest AGB in Guangxi is 988.17 Tg and the average forest AGB level is 73.30 t/ha. Natural broadleaf forest has the highest AGB level (86.75 t/ha), followed by natural coniferous forest (81.19 t/ha), planted coniferous forest (63.23 t/ha) and planted eucalyptus forest (49.71 t/ha). AGB level in karst areas is lower than that in non‐karst areas due to soil and water constraints. The majority of plantation forests in Guangxi is in the early and middle stages of forest succession, with the rapid growth of forest AGB, and has hence significant potential as carbon sinks. This study indicates that stack learning and incorporation of forest age could significantly decrease the uncertainty of forest AGB estimation. Our study helps to provide more accurate AGB information for karst ecological project management and regional carbon neutrality assessment.